mcp-searxng

What it does

The server integrates SearXNG—a privacy-first, self-hosted metasearch engine—into Claude and other AI assistants via MCP. It exposes web search, URL content extraction, search suggestions, and instance capability queries against any SearXNG instance you point it at. Results can be returned as formatted text or raw JSON, with filtering by language, time range, safe-search level, and relevance score.

Who it's for

Teams and individuals running private SearXNG instances who want Claude to perform web research within their own infrastructure—security teams, enterprises with data residency requirements, or privacy-conscious developers building AI tools that must not leak queries to third-party APIs.

Common use cases

• Research local news or internal content via a corporate SearXNG instance without exposing queries to cloud search providers.
• Ground LLM responses in real-time web results while keeping all traffic on-premise.
• Extract and process content from URLs discovered via search, with automatic caching to avoid redundant fetches.
• Autocomplete user search queries using SearXNG suggestions for faster exploration.
• Filter search results by language, date range, or safety threshold to meet compliance or content policies.

Setup pitfalls

• SEARXNG_URL must be reachable from the Node.js process; if behind a firewall or VPN, connection will silently fail. Test with curl first.
• SearXNG instances require explicit engine and category configuration; unknown names are rejected unless the /config endpoint is unreachable. Check your instance's available engines and categories first.
• The server writes filesystem cache for URL content; ensure the process has write permissions to its working directory.
• High risk class due to filesystem reads/writes and network calls—sandbox or run in a restricted environment if the instance is untrusted or handles sensitive data.